As an instrumentation engineer specializing in mining machinery, I have devoted my research to the health state analysis of power batteries for mining monorail vehicles. In the context of advancing mine safety, efficiency, and green intelligent construction, underground auxiliary transportation has become an integral component of coal mine production systems. The electric drive system of mining monorail cranes has progressively supplanted traditional diesel-driven equipment due to its operational convenience and zero-emission advantages. However, the safety and stability of the EV battery pack constitute the decisive factor governing the application and development of electric mining monorail systems. Consequently, accurately estimating the State of Health (SOH) of the battery and implementing effective monitoring and management mechanisms is of paramount importance. In this thesis, I investigate the health state analysis and management system construction for mining monorail power batteries, conducting the following research activities.
1. Aging Mechanism and Degradation Characteristics of Mining Monorail Power Battery
1.1 Working Environment Analysis of Mining Power Batteries
The operational environment for the EV battery pack in mining monorail systems presents substantial challenges. These challenges primarily manifest in hazardous and complex environments, spatial constraints, high operational intensity, maintenance difficulties, and performance requirements under specialized working conditions.
Underground environments contain flammable and explosive substances such as methane gas and coal dust. Consequently, the power battery enclosure must incorporate rigorous explosion-proof design alongside anti-seismic and impact-resistant features to prevent sparks from casing impacts. Moreover, the high-humidity conditions underground can lead to corrosion of the battery enclosure and internal circuits, necessitating excellent waterproofing and corrosion resistance for long-term reliable operation. The confined spaces of underground tunnels and chambers demand highly compact battery designs with small footprints and light weight for ease of installation and movement. However, the poor heat dissipation conditions underground allow heat to accumulate readily, necessitating efficient cooling systems to ensure stable operation.
The operational intensity represents another significant challenge for the EV battery pack. Underground auxiliary transport equipment typically requires continuous operation over extended periods, demanding high energy density from the battery to provide sustained and stable power output. Furthermore, the limited maintenance conditions in underground environments require the battery to possess high reliability and longevity, reducing failure rates while enabling essential maintenance procedures under constrained circumstances.
1.2 Basic Structure and Working Principle of Lithium Batteries
Lithium-ion batteries are secondary batteries characterized by high energy density, long cycle life, and no memory effect. Based on different cathode materials, they can be categorized into lithium iron phosphate (LiFePO₄), lithium cobalt oxide (LiCoO₂), and lithium manganese oxide (LiMn₂O₄) types. The choice of cathode material significantly determines battery performance including energy density, lifespan, safety, and cost. The anode material typically comprises graphite with high electrical conductivity and stability, whose layered structure permits lithium ion intercalation and deintercalation during charge-discharge cycles. Beyond electrode materials, lithium batteries consist of casing, electrolyte, and separator components.
The charging process primarily generates electric energy through Li⁺ migration between the cathode and anode. During discharge, Li⁺ migrates from the anode to the cathode, while during charging, Li⁺ returns from the cathode to the anode. Taking lithium iron phosphate battery as an example, which is widely used as the power source for mining electric monorail systems, the cathode is composed of LiFePO₄ while the anode adopts multi-layer graphite structure. The electrochemical reactions during charge-discharge can be expressed as follows:
$$Cathode: LiFePO_4 \rightleftharpoons Li_{1-x}FePO_4 + xLi^+ + xe^- \quad (2-1)$$
$$Anode: 6C + xLi^+ + xe^- \rightleftharpoons Li_xC_6 \quad (2-2)$$
$$Overall: LiFePO_4 + 6C \rightleftharpoons Li_{1-x}FePO_4 + Li_xC_6 \quad (2-3)$$
1.3 Factors Affecting Battery Performance Degradation
The external factors affecting the EV battery pack performance in mining monorail applications include mechanical abuse, temperature abuse, and electrical abuse. Mechanical abuse primarily manifests as strong mechanical impacts, continuous vibrations, collisions, and drops during operation. These factors can compromise battery performance and safety, with internal structure damage potentially leading to relative displacement between electrolyte and electrodes or even interface rupture. Temperature abuse occurs due to significant temperature fluctuations in underground environments, particularly in deep mines or high-temperature mining areas. Common high-humidity and dust-laden environments impede battery system heat dissipation, exacerbating heat accumulation. Electrical abuse results from frequent rapid acceleration and climbing demands that instantaneously elevate discharge rates, causing rapid cell temperature increase.
Internal degradation factors encompass cathode material aging, electrolyte decomposition, SEI film changes, anode lithium deposition, and self-discharge phenomena. During repeated charge-discharge cycles, the cathode material lattice undergoes volume expansion and contraction, with mechanical stress accumulating progressively. When stress exceeds thresholds, microcracks form on particle surfaces and propagate, eventually causing particle fracture. Concurrently, side reactions between cathode material and electrolyte generate non-conductive passivation layers that restrict further lithium ion insertion and extraction. SEI film, formed on the anode surface after initial charging, experiences rupture or thinning under extreme conditions, exposing anode material and intensifying side reactions. Anode lithium deposition occurs during charging when negative electrode lithium ion embedding capacity becomes limited, causing lithium metal to deposit on the anode surface in dendritic crystal form.
1.4 SOH Definition and Characterization Parameters
Battery State of Health serves as a metric for quantifying current capacity degradation, typically expressed as a percentage. Since battery SOH cannot be directly measured, indirect parameters such as capacity, internal resistance, charge, cycle count, and peak power are commonly utilized for estimation. The SOH can be defined from multiple perspectives:
From the perspective of battery capacity:
$$SOH = \frac{C_{now}}{C_{new}} \times 100\% \quad (2-4)$$
where C_now represents the current maximum available capacity and C_new denotes the rated nominal capacity.
From the perspective of battery internal resistance:
$$SOH = \frac{R_{EOL} – R_C}{R_{EOL} – R_{new}} \times 100\% \quad (2-5)$$
where R_EOL is the resistance at end of life, R_C is the current internal resistance, and R_new is the initial resistance of a new battery.
From the perspective of battery charge:
$$SOH = \frac{Q_{now}^{max}}{Q_{new}^{max}} \times 100\% \quad (2-6)$$
where Q_now_max represents the current maximum dischargeable charge and Q_new_max is the maximum discharge capacity from the factory.
From the perspective of remaining cycle count:
$$SOH = \frac{C_{remain}}{C_{all}} \times 100\% \quad (2-7)$$
where C_remain denotes the remaining charge cycles and C_all is the total rated charging cycles specified at the factory.
After comprehensive comparison and analysis, I determined that the capacity-based SOH estimation method is most practical and applicable, capable of achieving efficient estimation by combining partial charge-discharge data with intelligent learning algorithms. Therefore, I adopted this method for accurate and rapid battery SOH estimation in my work.
Common SOH characterization parameters for the EV battery pack include charge-discharge cycle count, voltage/current/time characteristics, internal resistance, and electrochemical impedance spectroscopy. Based on my comparative analysis of these characterization parameters, I chose to adopt multi-parameter fusion and comprehensive modeling based on different characteristic data to enhance the real-time accuracy, precision, and feasibility of SOH estimation for mining monorail power batteries.
2. Construction of SOH Estimation Model Based on SSA-PR-KELM
2.1 Data Source Analysis
Due to the difficulty of directly conducting charge-discharge tests on underground batteries and the current impossibility of obtaining complete experimental data from mining operations, I utilized the 18650 battery aging experimental dataset provided by the NASA Prognostics Center of Excellence for the health state analysis of mining monorail power batteries. The characteristics of 18650 cells, which constitute the EV battery pack in mining monorail systems, align closely with the health assessment standards of batteries in the NASA dataset.
I selected battery groups B5, B6, and B7 from the NASA dataset, which underwent 168 sets of discharge cycle tests at 24°C. The temperature conditions approximate the actual environment of 20°C to 40°C found in coal mine underground environments and align with the operating temperature of lithium batteries after cooling and explosion-proof treatment. Given the high degree of correspondence between the dataset and actual underground working conditions in terms of test temperature and health assessment standards, I determined this dataset to be a reliable data source for my research, providing robust data support for algorithm and model construction.
The charge-discharge testing procedure for B5, B6, and B7 batteries was as follows: batteries were first charged with a constant current of 1.5A. When the battery voltage reached 4.2V, the charging mode switched to constant voltage charging, during which the charging current gradually decreased. Charging ceased when the current diminished to 20mA. Subsequently, the batteries were discharged at a constant current of 2A until the voltage dropped to 2.7V, 2.5V, and 2.2V, respectively, whereupon the experiment concluded. These three battery groups possess more discharge cycles compared to other experimental groups, providing valuable experimental data for analyzing battery performance and health status. Table 1 presents the relevant parameters of the lithium-ion batteries.
| Battery ID | Constant Current Charge | CV Cut-off Voltage | CV Cut-off Current | Discharge Current | Discharge Cut-off Voltage | Temperature |
|---|---|---|---|---|---|---|
| B5 | 1.5A | 4.2V | 0.1A | 2.0A | 2.7V | 24.0°C |
| B6 | 1.5A | 4.2V | 0.1A | 2.0A | 2.5V | 24.0°C |
| B7 | 1.5A | 4.2V | 0.1A | 2.0A | 2.2V | 24.0°C |
2.2 Feature Extraction and Correlation Analysis
Since the internal structure of lithium-ion batteries, serving as the power source for mining electric monorail vehicles, continuously changes with increasing charge-discharge cycles, the actual available capacity decreases. Data such as voltage, current, time, and temperature during battery charge-discharge processes can reflect battery health status. To ensure these potential features adequately reflect battery health, I extracted eight health indicators (HI1-HI8) related to battery health from the lithium battery charge-discharge record data provided by NASA. Taking battery B5 as the research subject, I analyzed the relevant data as follows:
From the charging process voltage and current data at the 35th, 65th, 95th, 125th, 155th, and 168th cycles, I observed that with increasing cycle numbers, the time required for constant voltage charging significantly increased while constant current charging time diminished. In the voltage-time curve, the curvature change was greatest within the 3.9V to 4.2V range, while in the current-time curve, the range of 1.0A to 0.2A exhibited maximum curvature variation. I extracted constant voltage charging time as health indicator HI1, constant current charging time as HI2, the equal-voltage-rise charging time when terminal voltage ranged between 3.9V and 4.2V as HI3, and the equal-current-drop charging time when battery was in the 1.0A to 0.2A range as HI4.
From the discharge process voltage and temperature curves, I observed that with increasing cycles, both curves shifted leftward overall. I extracted the equal-voltage-drop discharge time when voltage was in the 3.8V to 3.3V range as HI5, and the time to reach maximum temperature during discharge as HI6. From the incremental capacity curves at different cycles, the peak value shifted rightward with increasing cycles, so I designated the incremental capacity curve peak as HI7 and the voltage corresponding to the IC curve peak as HI8.
To validate the effectiveness of the extracted health indicators, I employed Pearson correlation analysis to assess the correlation between HI1-HI8 and SOH. The Pearson correlation coefficient ranges from -1 to 1, where the closer the absolute value is to 1, the stronger the linear correlation between two variables. The calculation process is as follows:
$$\rho = \frac{\sum_{i=1}^{n}(x_i – \bar{x})(y_i – \bar{y})}{\sqrt{\sum_{i=1}^{n}(x_i – \bar{x})^2}\sqrt{\sum_{i=1}^{n}(y_i – \bar{y})^2}} \quad (3-1)$$
where ρ is the Pearson correlation coefficient, x_i and y_i represent the HI and SOH values respectively, and x̄ and ȳ represent the means of the health indicator and SOH values respectively.
Table 2 presents the correlation coefficients between each health indicator and SOH for batteries B5, B6, and B7. The results confirm that the selected health indicators exhibit strong correlation with SOH, thereby validating their effectiveness.
| Indicator | B5 | B6 | B7 |
|---|---|---|---|
| HI1 | -0.9106 | -0.9364 | -0.8841 |
| HI2 | 0.9961 | 0.9765 | 0.9934 |
| HI3 | -0.9106 | 0.9896 | 0.9884 |
| HI4 | -0.9644 | -0.9434 | -0.9509 |
| HI5 | 0.9991 | 0.9991 | 0.9996 |
| HI6 | -0.9363 | -0.8549 | -0.7617 |
| HI7 | 0.9873 | 0.9857 | 0.9792 |
| HI8 | -0.9061 | -0.9555 | -0.7506 |
2.3 Kernel Extreme Learning Machine and Hybrid Kernel Function
Extreme Learning Machine (ELM) is a learning algorithm for single-hidden-layer feedforward neural networks. Its network structure comprises an input layer, hidden layer, and output layer. Unlike traditional neural networks that employ back-propagation algorithms to iteratively adjust weights, ELM fixes the parameters from the input layer to the hidden layer and directly calculates the weights from the hidden layer to the output layer using the least squares method. This process does not require iterative optimization nor does it rely on complex quadratic optimization algorithms.
For a dataset containing N initial training samples (x_i, t_i), where x_i ∈ Rⁿ represents the n features of the input sample and t_i ∈ R^m denotes the label encoding of the sample, the ELM training process is expressed as:
$$\sum_{i=1}^{K} \beta_i g(a_i \cdot x_j + b_i) = y_j \quad (3-2)$$
where K is the number of hidden layer neurons, β_i is the weight from the i-th hidden neuron to the output neuron, a_i is the weight from the input layer neuron to the i-th hidden neuron, b_i is the bias of the i-th hidden layer system, g(x) is the infinitely differentiable activation function of hidden layer neurons, and y_j is the output value predicted by the network.
When the input error continuously decreases to zero, the following relationship holds:
$$\sum_{i=1}^{K} \beta_i g(a_i \cdot x_j + b_i) = t_j \quad (3-3)$$
In matrix form:
$$H\beta = T \quad (3-4)$$
where H is the hidden layer output matrix. When the number of hidden layer nodes is considerably smaller than the number of training samples, the output weights can be solved using the least squares method:
$$\beta = H^+ T \quad (3-5)$$
where H⁺ is the Moore-Penrose generalized inverse matrix of H.
Since the kernel matrix of the kernel extreme learning machine can substitute the random matrix HHᵀ, and the parameter I/C is added to the kernel function matrix to ensure non-zero eigenvalues, the output weight vector can be determined. The KELM model output is as follows:
$$f(x) = \begin{bmatrix} K(x, x_1) \\ \vdots \\ K(x, x_N) \end{bmatrix}^T \left(\frac{I}{C} + \Omega_{ELM}\right)^{-1} T \quad (3-6)$$
To achieve superior performance, I combined the local and global kernel functions to form a hybrid kernel, specifically linearly weighted combination of polynomial kernel and Gaussian kernel:
$$K_{PR}(x_i, x_j) = b \cdot K_P(x_i, x_j) + (1-b) \cdot K_R(x_i, x_j) \quad (3-7)$$
where K_PR is the hybrid kernel function, K_P is the polynomial kernel function, K_R is the Gaussian kernel function, and b is the balancing factor. This hybrid kernel function combines the advantages of both local and global kernels, providing excellent learning and generalization capabilities.
2.4 Sparrow Search Algorithm
The Sparrow Search Algorithm (SSA) is a novel swarm intelligence optimization algorithm designed by mimicking sparrow foraging and predator avoidance behaviors. The sparrow population is broadly divided into two types: explorers with higher fitness values and followers. During the process, followers decide whether to follow explorers based on fitness value changes and continuously update and correct their positions according to specific rules.
The position update of explorers is governed by:
$$X_{i,j}^{t+1} = \begin{cases} X_{i,j}^t \cdot \exp\left(-\frac{i}{\alpha \cdot T}\right) & \text{if } R < ST \\ X_{i,j}^t + Q \cdot L & \text{if } R \geq ST \end{cases} \quad (3-8)$$
where t represents the current iteration number, T is the maximum iteration count, R ∈ [0,1] is the warning value, ST ∈ [0.5,1] is the safety value, Q is a random number following normal distribution, and L is a 1×d matrix with all elements equal to 1.
The follower position update is:
$$X_{i,j}^{t+1} = \begin{cases} Q \cdot \exp\left(\frac{X_{worst}^t – X_{i,j}^t}{i^2}\right) & \text{if } i > n/2 \\ X_P^{t+1} + |X_{i,j}^t – X_P^{t+1}| \cdot A^+ \cdot L & \text{if } i \leq n/2 \end{cases} \quad (3-9)$$
The parameter settings for the SSA algorithm used in my work are presented in Table 3.
| Parameter | Value |
|---|---|
| Initial population N | 30 |
| Maximum iterations Max_iter | 250 |
| Producer proportion PD | 0.2 |
| Scout proportion SD | 0.8 |
| Warning threshold ST | 0.6 |
2.5 Model Establishment Process for SOH Estimation
Based on the efficient learning capability of the hybrid kernel extreme learning machine and the advantages of SSA in global optimization, I proposed a SSA-optimized hybrid kernel extreme learning machine model. The specific process is as follows:
First, I extracted health indicators from the B5, B6, and B7 lithium battery datasets, including constant voltage charging time, constant current charging time, equal-voltage-rise charging time, equal-current-drop charging time, equal-voltage-drop discharge time, time to reach maximum temperature during discharge, IC curve peak value, and IC curve peak corresponding voltage. After confirming their correlation with SOH through Pearson correlation analysis, I selected these eight health indicators as model input data.
Second, I preprocessed the data, using the first 70% as the training set and the remaining 30% as the test set for validation of model prediction performance.
Third, I applied the SSA algorithm to the parameter optimization process of the PR-KELM model, setting the population size to 30 and the number of iterations to 250, selecting RMSE as the fitness function:
$$RMSE = \sqrt{\frac{1}{m}\sum_{i=1}^{m}(y_i – \hat{y}_i)^2} \quad (3-10)$$
Fourth, I employed the SSA-PR-KELM estimation model to predict the SOH of the test sets of B5, B6, and B7 lithium batteries respectively. I then compared the predicted results with the true SOH values and calculated the errors between them to verify the capability of the SSA-PR-KELM model in estimating lithium battery SOH.
2.6 Estimation Results and Comparative Analysis
Using the extracted health indicator data as model input and battery SOH as output, I evaluated the model on three lithium batteries. The maximum estimation error for B5 occurred at the 151st cycle, reaching 0.436%, with the model exhibiting slight delay at capacity regeneration boundaries. For B6, the maximum error reached 0.78% at the 152nd cycle. For B7, the maximum error was 0.23% at the 128th cycle. Overall, the SOH estimation errors for all three batteries remained within acceptable ranges.
To demonstrate the superiority of the SSA-PR-KELM model in assessing SOH accuracy, I employed three evaluation metrics: coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The evaluation metrics are calculated as follows:
$$MAE = \frac{1}{m}\sum_{i=1}^{m}|y_i – \hat{y}_i| \quad (3-11)$$
$$RMSE = \sqrt{\frac{1}{m}\sum_{i=1}^{m}(y_i – \hat{y}_i)^2} \quad (3-12)$$
$$R^2 = 1 – \frac{\sum_{i=1}^{m}(y_i – \hat{y}_i)^2}{\sum_{i=1}^{m}(y_i – \bar{y}_i)^2} \quad (3-13)$$
I also established P-KELM, SSA-P-KELM, SSA-R-KELM, and SSA-L-KELM models for comparative analysis. Table 4 presents the comparative analysis of the prediction results of each algorithm model.
| Battery | Algorithm Model | R² | RMSE | MAE |
|---|---|---|---|---|
| B5 | P-KELM | 0.9912 | 0.0036 | 0.0028 |
| SSA-P-KELM | 0.9800 | 0.0056 | 0.0046 | |
| SSA-R-KELM | 0.9878 | 0.0044 | 0.0033 | |
| SSA-L-KELM | 0.9791 | 0.0057 | 0.0047 | |
| SSA-PR-KELM | 0.9927 | 0.0034 | 0.0026 | |
| B6 | P-KELM | 0.7758 | 0.0315 | 0.0243 |
| SSA-P-KELM | 0.9947 | 0.0049 | 0.0036 | |
| SSA-R-KELM | 0.8814 | 0.0230 | 0.0188 | |
| SSA-L-KELM | 0.9757 | 0.0104 | 0.0086 | |
| SSA-PR-KELM | 0.9941 | 0.0051 | 0.0038 | |
| B7 | P-KELM | 0.9769 | 0.0058 | 0.0045 |
| SSA-P-KELM | 0.9966 | 0.0022 | 0.0018 | |
| SSA-R-KELM | 0.9906 | 0.0037 | 0.0028 | |
| SSA-L-KELM | 0.9967 | 0.0023 | 0.0018 | |
| SSA-PR-KELM | 0.9966 | 0.0022 | 0.0018 |
From the comparative analysis, I observed that on B5, B6, and B7 batteries, the SSA-P-KELM model significantly outperformed P-KELM. The SSA-PR-KELM model demonstrated superior prediction accuracy on the B5 and B6 batteries compared to the SSA-optimized single-kernel extreme learning function models, and comparable performance on B7. The experimental results show that the SSA-PR-KELM model achieves estimation errors reduced by 15% to 22% compared to other models, with maximum errors not exceeding 0.78%, demonstrating superior estimation accuracy and robustness.
3. Hardware Design of the Health Management System for Single-Rail Power Batteries
3.1 Overall Hardware Architecture
The battery health management system for electric monorail cranes is designed to collect and monitor characteristic parameters of lithium batteries in underground environments, displaying relevant data intuitively to operators. Upon detection of abnormal battery conditions, the system immediately issues alerts, enabling underground personnel to take appropriate emergency measures and thereby avoiding potential safety hazards during underground battery operation.
The hardware system architecture encompasses the STM32 microcontroller as the core processing unit, complemented by data acquisition modules including differential amplifier circuits for voltage measurement, Hall-effect sensors for current detection, NTC thermistors for temperature monitoring, MPU-6050 sensors for acceleration measurement, NB-IoT wireless communication modules, and power supply circuitry providing stable voltage rails at 12V, 5V, and 3.3V through multi-stage linear regulation.
3.2 Main Control Chip Selection and Circuit Design
I selected the STM32F103R6T6 from STMicroelectronics as the main control chip, which provides rich peripheral resources, high stability, and rapid computation capabilities. The STM32F103R6T6 operates at a maximum frequency of 72MHz in an LQFP64 package, supports various memory types, and functions within an output voltage range of 2.0V to 3.6V. It includes two independent 12-bit analog-to-digital converters, three general-purpose 16-bit timers, one advanced timer for PWM control, two SPI interfaces, three USART interfaces, two I²C interfaces, and one CAN bus interface. The chip operates normally within a temperature range of -40°C to 85°C, with high-speed embedded memory capacity reaching 512K bytes.
3.3 Data Acquisition Module Design
For voltage acquisition, I employed a differential amplification approach. Considering the special mine environment that may cause signal interference from external noise, temperature variations, or other common-mode signals, I designed the voltage acquisition circuit using differential amplification to measure the voltage difference between input signals while eliminating common-mode interference. The circuit utilizes an LM358 operational amplifier configured in two stages: the first stage performs differential amplification to extract the voltage difference and reject common-mode noise, while the second stage provides additional amplification to ensure adequate signal levels for subsequent analog-to-digital conversion.
The current sensing element employs the ACS724 Hall-effect current sensor from Allegro MicroSystems, which offers excellent linearity and low noise characteristics. The sensor is capable of bidirectional measurement up to ±5A with a sensitivity of 200mV/A. The output voltage relates to the measured current as follows:
During discharge: V_OUT = 2.5V + 0.2V/A × I
During charging: V_OUT = 2.5V – 0.2V/A × I
For temperature monitoring, I selected the MF52D103F3950 NTC thermistor with a water-drop temperature measurement configuration. This thermistor operates over the temperature range of -30°C to +105°C, fully conforming to the operating requirements of mining monorail power batteries. The thermistor is configured in a voltage divider circuit to convert temperature-dependent resistance changes into measurable voltage signals:
| Temperature (°C) | Resistance (Ω) | Temperature (°C) | Resistance (Ω) |
|---|---|---|---|
| -40 | 336600 | 19 | 13070 |
| -39 | 315000 | 20 | 12490 |
| -38 | 295000 | 21 | 11940 |
| -37 | 276400 | 22 | 11420 |
The system also incorporates a 0.96-inch OLED display with 128×64 resolution using I²C interface communication and SSD1306 driver chip, providing real-time data visualization with low power consumption and high contrast. For vibration monitoring, I employed the MPU-6050 sensor, which integrates a three-axis accelerometer and three-axis gyroscope with high-precision 16-bit ADCs. The sensor supports I²C and SPI communication interfaces and operates within a temperature range of -40°C to +105°C. The acceleration data is derived by dividing the raw values by the sensitivity factor (2048 LSB/g at ±16g range).
3.4 Wireless Communication Technology Analysis
Through comparative analysis of various communication technologies, I selected NB-IoT for the system due to its excellent penetration capability, superior coverage range, low power consumption, and stability in underground environments. Unlike Bluetooth, Wi-Fi, and ZigBee which may experience signal attenuation and coverage deficiencies in underground environments, NB-IoT provides reliable wide-area coverage and can effectively penetrate complex underground environmental conditions.
| Parameter | Bluetooth | Wi-Fi | ZigBee | LoRa | NB-IoT |
|---|---|---|---|---|---|
| Data rate | 1-3Mbps | 11-45Mbps | 20-250Kbps | 0.3-50Kbps | 100-250Kbps |
| Coverage | 20-200m | 20-200m | 2-20m | 1-20km | >10km |
| Frequency band | 2.4GHz | 2.4GHz | 2.4GHz | Sub-GHz | Licensed band |
| Power consumption | Low | High | Low | Low | Low |
| Security | High | Low | Medium | High | High |
I selected the BC26 module for wireless communication implementation, which offers excellent performance and high reliability, ensuring stable operation and communication quality in complex underground environments.
4. Software Design of the Health Management System
4.1 Software Development Environment
I utilized Keil uVision5 as the software development platform, which provides comprehensive development solutions including extensive integrated development and debugging tools, complete function call libraries, and enhanced design efficiency. The platform allows direct programming, editing, and debugging of various program modules following target chip selection, project initialization, compiler configuration, and library file addition.
4.2 System Operating Flow Design
The health management system is designed to collect, analyze, and display battery operational data in real time through a modular interface. After system initialization and hardware setup, the main loop executes data acquisition, processing, display, and communication functions. Real-time monitoring of sensor data includes threshold-based checking with alarm notification through buzzer and OLED display systems. Under normal operating conditions, data is transmitted to the OneNET platform through the NB-IoT module while the OLED display updates in real time.
4.3 Data Acquisition Module Programming
The voltage acquisition program flow begins with battery signal input through the differential amplifier configuration, followed by low-pass filtering to remove high-frequency noise, analog-to-digital conversion by the built-in ADC, and final voltage value output. For current acquisition, the Hall sensor output voltage is digitized and converted using the sensitivity calibration equations described earlier. The temperature measurement program utilizes the NTC thermistor lookup-table method, converting resistor values to temperature readings, while the MPU-6050 acceleration acquisition flow performs signal acquisition, A/D conversion, digital filtering, and data storage.
4.4 SOH and SOC Estimation Software Design
I implemented a combined ampere-hour integration and open-circuit voltage method for SOC estimation in the software design. The initial SOC value is determined by measuring open-circuit voltage when the battery is in a stable state, whereas during battery operation, SOC is estimated through current integration:
$$SOC = SOC_0 + \frac{1}{Q_N}\int_{0}^{t} I(\tau) d\tau \quad (4-1)$$
where SOC₀ is the initial state of charge, Q_N is the nominal capacity of the battery, and I(τ) is the instantaneous current.
The system also acquires and analyzes battery charge-discharge data to extract health factors. Depending on the need for new datasets, the system re-trains the health state estimation model or directly utilizes the trained model for SOH estimation. The hybrid approach combining static open-circuit voltage and dynamic current integration provides comprehensive and accurate SOC estimation, adapting to the dynamic characteristics of batteries under various working conditions.
4.5 NB-IoT Module Access and Configuration
The BC26 module configuration involves AT commands to establish communication with the cloud platform. The main AT commands used for configuration are presented in Table 7.
| AT Command | Function Description |
|---|---|
| AT | Power-on check and communication module response test |
| AT+CIMI | Verify SIM card reading success |
| AT+CESQ | Check signal strength and refresh latest signal status |
| AT+CFUN | Module function mode setting |
| AT+CGATT=1 | Device access to mobile network |
| AT+CGATT | Verify module network connection status |
| AT+CSQ | Acquire current device signal strength |
| AT+NRB | Restart the module |
During system configuration, the STM32 microcontroller sends AT commands through the serial port to configure the NB-IoT module’s MQTT connection parameters. This configuration enables the module to establish connection parameters with the MQTT server, including device ID, secret key, service address, and other essential cloud platform authentication elements. Following successful configuration, the module initiates a login request to the cloud platform through network connection, establishing a communication channel for data upload and command reception.
4.6 Cloud Platform Integration and Implementation
I selected OneNET, a comprehensive IoT platform from China Mobile, as the system’s cloud platform due to its capabilities in device access, data storage, machine control, and protocol adaptation. OneNET supports multiple communication protocols including LWM2M, Modbus, and HTTP, while providing one-stop device management services. The platform integration process involves the following steps:
First, account registration and login through the OneNET official website. Second, product creation in the developer center, selecting the NB-IoT IoT suite as the product type with LWM2M as the protocol. Third, device addition with automatic provisioning of unique device identification codes for ensuring device uniqueness. Fourth, data stream template configuration with parameters including data stream names, data units, and data types, ensuring consistency between device-side programs and cloud platform configurations. Fifth, device connection and data reporting where the BC26 module interacts with the OneNET cloud platform to establish online connection status and verify real-time data stream updates.
4.7 Host Computer Platform Design
The host computer monitoring platform is designed to receive, process, display, and analyze data transmitted from the lower computer. I employed C++ language programming in the QT development environment for the upper computer interface design. MySQL serves as the database management system for data storage and management, providing efficient, reliable, and secure data storage solutions for the monitoring system.
The database design uses DBeaver, an open-source universal database management tool, which provides a unified interface for connecting and managing database systems. Through database parameter configuration, connection establishment, and data table creation, the system achieves efficient data organization and management for subsequent visualization and analysis.
The upper computer interface includes a user authentication module for system security, requiring valid account and password credentials for login access. The main display interface presents real-time battery operational data including individual cell voltages, working temperatures, total current, vibration accelerations, SOC and SOH values. The interface incorporates multiple data visualization modules with real-time curve displays for temperature, voltage, current, and acceleration parameters, enabling operators to monitor battery working states effectively. The threshold configuration functionality allows staff to adjust alerting parameters based on actual operational requirements.
5. Comprehensive Experimental Verification
5.1 Experimental Platform Construction Based on Similarity Theory
Based on similarity theory principles, I constructed an experimental testing platform using 18650 cylindrical lithium iron phosphate batteries. Although these batteries differ in physical dimensions and packaging forms from actual mining monorail power batteries, they exhibit high similarity in key performance parameters including electrochemical characteristics, thermal characteristics, and load response behaviors. The selected batteries support high-rate discharge and demonstrate good temperature stability and cycle life, making them suitable for establishing a representative experimental model.
The system design fully referenced the actual working conditions that mining monorail systems may face in operation, analyzing battery working state response characteristics to establish a foundation for subsequent application verification and functional expansion under complex working conditions.
5.2 System Functional Testing
The hardware testing procedure involved pre-welding connection of modules using Dupont wires and initial power supply verification to ensure normal operation of all modules. After confirming correct module functionality, soldering to the circuit board was performed with careful attention to avoid soldering defects. Following soldering, electrical parameter measurement using multimeters verified connection points and circuit integrity. After completing hardware inspection, the system underwent functional testing to verify data acquisition and communication performance.
The system testing platform included a computer, data cables, terminal nodes, and gateway nodes. The PC was equipped with MDK development environment and debugging tools for convenient debugging and data inspection. USB serial port connection established communication with the PC, using a serial port debugging assistant and cloud platform for real-time data transmission monitoring. The NB-IoT wireless communication module was tested using the serial port detection tool with COM3 port settings including baud rate of 115200bps, 8-bit data, and 1-bit stop.
5.3 Cloud Platform Testing Results
The cloud platform served as the intermediate link between the terminal system and the upper computer. Device connection status verification was performed through the device list interface in OneNET, confirming successful network connection of the NB-IoT module. Data stream confirmation ensured accurate transmission of collected information from the communication module. Real-time monitoring of the EV battery pack data was achieved through the platform’s data display interface.
5.4 Host Platform Testing Results
I conducted comprehensive testing of the upper computer platform system. The system retrieves real-time data from the cloud platform, stores it in the database, and displays it on the remote interface. The battery state display uses color coding for intuitive understanding: green indicates battery charge remaining 60%-100% with good health status, yellow indicates 20%-60% battery charge remaining with acceptable health status, and red indicates both battery remaining charge and health status below 20%, alerting maintenance personnel to perform timely battery maintenance.
Real-time change curves for temperature, voltage, current, and acceleration intuitively reflect battery operating states under different working conditions. When parameter curves continuously rise to set thresholds, the system automatically triggers alarms and disconnects the load, while preserving historical records in the database for subsequent analysis and maintenance strategy optimization.
6. Conclusion and Future Prospects
In this thesis, I conducted comprehensive research on health state analysis and management system for power batteries in mining monorail vehicles. Through systematic investigation and experimental validation, I achieved the following research outcomes:
First, I thoroughly analyzed the challenges faced by lithium batteries in underground environments, dividing factors affecting battery health into internal aging factors and external environmental factors. I established battery health state definitions from multiple perspectives including capacity, internal resistance, charge, and remaining cycle count, providing a theoretical foundation for subsequent battery health management.
Second, I proposed a lithium battery SOH prediction method based on indirect health factors. By extracting health indicators from public datasets and employing the SSA-PR-KELM model optimized by sparrow search algorithm for battery SOH prediction, I demonstrated that the model exhibits superior accuracy and applicability compared to other prediction models, effectively reducing prediction errors and showing good robustness across different batteries in the dataset.
Third, I designed and implemented a comprehensive health management system for the EV battery pack, including main control, data acquisition, OLED display, and communication modules. The system achieves remote management of power battery health status. Through the establishment of a test platform using 18650 lithium batteries with properties similar to mining monorail power batteries, and the integration of NB-IoT modules for wireless data transmission, I verified the normal operation and real-time display and analysis of battery status.
Fourth, I conducted comprehensive functional tests and hardware debugging to ensure normal operation of all functional modules and validate system stability and reliability. Experimental results confirmed that the system executes predetermined functions stably and meets design requirements, providing reliable technical support for subsequent practical applications.
Future research directions may include: employing actual mining electric monorail power battery charge-discharge data combined with additional intelligent algorithms to improve model adaptability and SOH prediction accuracy; and improving the mechanical structure of the mining monorail power battery enclosure through new shell materials, enhanced vibration protection, and optimized heat dissipation systems to improve water resistance, dust prevention, and seismic performance.